DS808: Visualization project

By Josephine Plass-Nielsen and Ulrikke Jørgensen

Domain situation

The National Center of Environmental Information has the biggest archive of registred earthquakes along with features as:

• Country
• Lattidude
• Longitude
• Focal depth
• Year
• Magnitude
• Total number of death
• Total number of missing
• Total number of injuries
• Total number of houses damaged
• Total number of houses destroyed

When starting a project like this, one will have to look at the domain situation to be able to generalize the data and tasks into something more generic. This includes the task of translating domain-specific language into more abstract and proper form, so it will be easier to reason about similarities and and differences. The data we are working with comes from the National Center of Environmental Information which monitors natural hazards.

Target user

Our target user is ordinary (non-expert) people that are interested in knowing more about earthquakes.
An example could be a person who wants to see the distribution of earthquakes in the world, or wants to discover correlations between the features of earthquakes.

To be able to abstract the data and tasks in a way that makes sense, one will have to determine the target users. In our case, the target user will be non-expert people looking to see patterns in regards to some preferred attributes of the earthquakes. This could be to see the distribution of earthquakes on the world or perhaps interesting correlations.

WHAT - Data abstraction

Dataset type --> table

Data types --> items & atributes

Availability --> dynamic/static

Atribute types -->

data 

	Year	Country	Latitude	Longitude	Focal Depth (km)	Mag	Total Deaths	Total Missing	Total Injuries	Total Houses Destroyed	Total Houses Damaged
0	1935	USA	46.600	-112.000	0	6.2	2	0	0	300	0
1	1935	CHINA	29.400	102.300	0	6.0	2	0	0	0	0
2	1935	USA	46.600	-112.000	0	6.0	2	0	0	0	0
3	1935	CHINA	28.700	103.600	0	6.0	100	0	0	0	0
4	1935	TAIWAN	24.600	120.800	30	6.5	2746	0	6004	30000	0
...	...	...	...	...	...	...	...	...	...	...	...
2503	2020	USA	40.751	-112.078	12	5.7	0	0	0	0	0
2504	2020	NEW ZEALAND	-33.294	-177.838	10	7.4	0	0	0	0	0
2505	2020	CHINA	33.124	98.916	10	5.3	0	0	0	0	790
2506	2020	INDONESIA	-6.808	106.676	23	5.0	0	0	4	0	1137
2507	2020	USA	55.030	-158.522	28	7.8	0	0	0	0	0

2508 rows × 11 columns

Country er categorical og ellers er resten Quantitative

Longitude and Latitude er diverging og Total Deaths er sequential

Tilsammen giver Long og Lat en position

What conclusions can be derived from the data?

corrMatrix = data_sub.corr()
sn.heatmap(corrMatrix, annot=True)

<matplotlib.axes._subplots.AxesSubplot at 0x237ee4769c8>

So what conclusions can be derived from the data? To know that we have made a correlation matrix to see if there is any interesting correlations for our target users. In the matrix, we can see that the strongest correlations are between number of deaths and number of injuries which is quite obvious although not so interesting. Also, there is perhaps a weak correlation between number of injuries and number of houses destroyed. Interestingly, there is almost no correlation between the number of magnitude and the number of deaths. Therefore, we check the distribution of the attributes too.

x = pd.plotting.scatter_matrix(data_sub, diagonal='hist', figsize=(14,10))   
for ax in x.ravel():
    ax.set_xlabel(ax.get_xlabel(), fontsize = 7.5, rotation = -15); ax.set_ylabel(ax.get_ylabel(), fontsize = 8, rotation=83)

As it can be seen in the large scattermatrix, most of the distributions are uniform with except from some outliers. However, one thing that we found interesting was that it looks like there has been more earthquakes in recent years.

plt.scatter(data_sub['Year'],data_sub['Mag'])
plt.ylabel('Mag')
plt.xlabel('Year')
plt.show()

Also, along the years it seems the the outliers of deaths get more extreme.

plt.plot(data_sub['Year'],data_sub['Total Deaths'])
plt.ylabel('Total Deaths')
plt.xlabel('Year')
plt.show()

WHY - Task abstraction

Action and target:

• Locate exstremes

• Summarize distribution

• Why do i want to use visualisation for my problem?

• Do the visualization support the underlining task?

Moving to the why question and the task abstraction.

To specifi why we want to use visualisation, we fist need to define our action and target.

We started out without a plan for our visualization. We therefore, as shown before, started out at a midlevel action, where we explored our dataset of earthquakes, with the perpose of getting an overview, to see what appears and if patterns where formed, the goal was to deriving our target, which can be special features, outliers or trents, that we can work forward with at a higer level of action.

As shown before we found some exstremes in magnitude Total Death, we also found an interesting destribution in earthquakes over the years. Moving forward our action and target therefore becomse: locate exstremes and summarie destribution.

So, to answer why we want to use visualisation: a visualisation support our underlinging task well, because we are dealing with data that is registred trough time and also by place, visualising the data on a map trough a timeline is therefore a very represntative presentation for time and place.

HOW - Visual encoding/interaction idiom

Which design decisions are taken?

Mapping data features to visual features:
Glyph map: Earthquake = glyph
Place (latitude, longitude) = World map
Number of death = color of glyph
Magnitude = size of glyph
Year = timeline

From magnitude to realistic size of glyph

The fist design desision we have to make, is how we map our data features to visual features.

We choose to map our data as a glyph map: the glyph map draws a glyph for each data point and uses the visual features color, shape and size to represent data features.

In our case Each glyph on the map represents an earthquake, the color the number of deaths ranging from dark red to light yellow indicating how deadly the earthquakes was, the size of the glyph represents the magnitude indicating the size of an earthquake.

The disadvantage of a glyph map can be that it is limited to a certain amount of data, as when the number of glyphs increases, it gets harder to distinguish the observations from each other. This is a factor that we take into account coding the sizes of the glyphs. however this is also a problem that is solved by plotting our data on a timeline spreeding the observations out.

The glyph map also allows us to code a hover function, which means that the user can hover over a data point to see the data for the given observation. The function allows us to present more data at the same time.

Cartography: For the map, we have chosen a Robinson liked styled map. The Robinson styled map gives a good representation of the world, because it does not compromise projection and it reduces distortion of land masses near the poles, which can be seen in other maps. we therefore chosse this map as it is a good representation of how we perceive the world, and because we are not mapping any distances this is the most important factor.

As we are handeling a quite big data set, we have choosen to do some filtering to simplyfy our visualisation. We have choosen to take obsevasions from year 1935 and to 2020. The scale of magnitude was invented at that time and so we belive that the observations here after will be more relierble.

We have also chossen to focus on the biggest earthquakes with a magnitude of 5 and above, to focus on the largest earthquakes.

Filtering data:

• From year 1935 and after
• From magnitude 5 and above

So, as an conclution: from our visualisation, outliers becomes very visually, where these big earthquakes really striks out visually.

Also the destributions of earthquakes trough time becomse more visual on the timeline, althoug is is a bit hard to see the increasing in number of earthquakes ??? ## hvad tænker du? Tror måske, at det er bedre at sige det til eksamen :)

fig = px.scatter_geo(data_test, lat=data_test["Latitude"], lon=data_test["Longitude"], color=data_test["Total Deaths"], hover_name=data_test["Country"], hover_data={'Latitude': False,'Longitude': False,'Bubble Size': False,'Year': True,'Mag': True,'Total Deaths': True}, size=data_test['Bubble Size'],size_max = 60, color_continuous_scale=["orange", "red", "brown", "black"], range_color=(0,316000), opacity=1, animation_frame=data_test['Year'], animation_group=data_test["Year"], projection="hammer")
fig.show()